Research Information System
VH-Roda 2024, Rome, Italy, 2 - 06 December 2024, pp.1-2, (Summary Text)
The Ground Control Points (GCP) are involved in geometric Calibration / Validation activities of remote sensing images and used as reference measurements. An observation of yearly VH RODA workshops is that even if GCP dataset plays a crucial role in data processing and data quality control, their availability as free and open-source data to a broad community remains reduced, specifically for very high spatial resolution data. Thanks to the GCPIx initiative (Strobl, 2023), the Committee on Earth Observation Satellites (CEOS) is now proposing the development of a harmonised global CEOS GCP Database and its extension to cover also VHR Optical Data [2.5-10m GSD, and potentially 2.5m GSD]. Existing tools and demonstration services pooled by NASA, ESA, Australia government, as illustrated by Saunier et al. (2023), are already fully available under specific agreement for some of them. However, today, in quite a few geographical areas VHR GCPs does not exist, and the expansion of this database remains a long process of discussions. Besides, tailoring classic concept of GCP database for VHR still faces to some barriers mostly due to on the one hand data sharing agreement (Copyright / Licensing), and on the other hand harmonization (GCP description) and uncertainties due to radiometric and geometric differences between input and reference data. In this poster presentation, we plan to discuss a new concept of GCP database which rely on the generation of GCP image chips based on Artificial Intelligence, in particular generative adversarial networks (GANs). This concept benefits from recent developments and major achievements sourced from Deep Learning (DL) methods. In addition, the approach account for the massive quantity of VHR and Unmanned Aerial Vehicle (UAV) data that can be nowadays found freely available but improvements to their geometric quality are often needed. After describing the concept, the poster first discusses the algorithm approach, in particular the training dataset preparation, the selection of the DL architecture, and the DL hyperparameter tuning. Finaly, the first results obtained from the generation of the GCP image chips with 10 m and 50 cm resolution are reported.